Hermes Agent, the open-source AI agent runtime from Nous Research, has been steadily evolving. Version 0.11 introduces several significant improvements for developers building agentic applications.
Enhanced MCP Integration
The most notable update is deeper integration with the Model Context Protocol (MCP). Hermes now supports:
- Native MCP client for connecting to any MCP-compatible server
- Auto-discovery of tools and resources from MCP servers
- Seamless fallback when MCP servers are unavailable
This makes Hermes one of the most MCP-compatible agent frameworks available. For more on MCP, read our deep dive on the protocol.
New Toolsets
Version 0.11 introduces a refined toolset system that lets developers restrict agent capabilities more granularly:
- Browser toolset — Headless browser navigation and interaction
- Terminal toolset — Secure shell command execution
- File toolset — Read/write operations with path restrictions
- Web toolset — HTTP requests and web search
- Vision toolset — Image analysis capabilities
Each toolset can be enabled or disabled per-agent, giving precise control over what each agent can access.
Multi-Model Routing
Agents can now be configured to use different models for different tasks:
agent:
reasoning_model: deepseek-v4-pro
code_model: claude-sonnet-4
vision_model: gpt-4o
This lets developers optimize for cost and capability per-task rather than using a single model for everything.
Getting Started
Hermes Agent runs on any Linux machine and can be installed via pip:
pip install hermes-agent
hermes setup
The project is fully open-source under the MIT license. Check the documentation and GitHub repo for more details. For broader context on the open-source agent ecosystem, see our ultimate guide to agent frameworks.